Logging based Coordinated Checkpointing in Mobile Distributed Computing Systems

Abstract

Checkpointing is an efficient way of implementing fault tolerance in distributed systems. Mobile computing raises many new issues, such as high mobility, lack of stable storage on mobile hosts (MHs), low bandwidth of wireless channels, limited battery life and disconnections that make the traditional checkpointing protocols unsuitable to checkpoint such systems. Checkpointing can be independent, synchronous, quasi-synchronous, or message logging based. In synchronous checkpointing, all or interacting processes need to checkpoint synchronously, extra synchronization messages are sent, some information may be piggybacked onto computation messages, blocking of processes may take place, and in case of fault, all processes are forced to rollback. It becomes difficult for multiple MHs to checkpoint synchronously due to disconnections and unreliable wireless channels. MHs are prone to frequent failures, which will require frequent rollback of all processes. In this paper, we propose a hybrid non-intrusive checkpointing protocol, where fixed hosts checkpoint synchronously and MHs checkpoint independently. The proposed scheme gives MHs autonomy in taking checkpoints and reduces the information to be piggybacked onto computation messages. An MH can recover independently by using its recent checkpoint and message log without forcing other nodes to rollback.